AI in a law firm: the claim goes to the chat, client data stays in the firm

What happens in the film

Tuesday, 6:20 p.m. Jake, a trainee, is finishing a claim and wants the AI to shorten the reasoning and check the dates for consistency. He sends the file to an AI chat. The file holds the plaintiff's data: national ID number, address, phone number, bank account number. In the margin there is a comment left over from the template, with the name and address of a client from the previous case.

The partner reminds him of professional secrecy and that the data of two clients is involved. The data protection officer says what comes next: an entry in the register, a risk assessment, and the decision is the firm's. The evening ends with a phone call to the client.

Take two, this time with the extension. The same claim, the same move. The program lists the detected data of both clients, the comment included. The chat gets a copy with replacements. The reply comes back with the real data, and Jake checks it sentence by sentence and leaves on time.

What the program detects in a law firm's documents

The detection rules are tuned for Polish data: PESEL (the Polish national ID number), NIP (the Polish tax ID), REGON (the Polish business registry number) and Polish addresses and IBAN account numbers. Names, e-mails and phone numbers are also found in English text.

• first names and surnames, also a bare surname later in the text and inflected forms (Polish case endings);
• national ID numbers, tax IDs, business registry numbers, bank account numbers, ID card numbers, phone numbers, e-mails, addresses and postal codes with the city;
• company names with a legal form (for example "sp. z o.o." or "S.A.");
• in docx files: the text of the document, headers, footers, footnotes, comments and hidden text. The chat gets a clean copy rebuilt from the content alone (comments as plain text), without formatting, images and macros.

What the program doesn't change, or may miss:

• case reference numbers, land and mortgage register numbers, court names, dates and amounts;
• company names written without a legal form, and a person who appears only once and only in an inflected form: add such names as custom phrases to mask;
• the description of the facts. The program replaces identifying data, but the facts of the case stay, because the AI works on them. Whether such a description may go to a tool is for the lawyer to judge;
• PDF files and scans: they are stopped by default, not masked. Old .doc files and password-protected files are stopped too.

What the rules say

Lawyers are bound by professional secrecy obligations that cover everything they learn in connection with legal advice. The secrecy covers the facts of the case, not only personal data. Professional bodies may add their own guidance on using AI: the tool helps, the lawyer verifies the result, and secret information doesn't go into tools that don't offer adequate protection. These rules differ from country to country, so check the ones that apply to you.

For its clients' data the firm is a data controller under GDPR. It records every breach in a register (Art. 33(5)), assesses the risk and, where needed, reports it to the data protection authority, where feasible within 72 hours (Art. 33(1)). If the risk is high, it also notifies the clients (Art. 34). GDPR names pseudonymisation as an example of a technical measure (Art. 25 and 32), but data with replacements are still personal data (recital 26).

Questions and answers